MLP model question

[frabul]

Hi, i have read the paper and i gave a fast glance at the code.
I would like to ask how was possible to for a feed forward network (MLP) to modelize a dynamic system? As far i know MLP has no memory capabilities that means it's output is directly dependent on inputs only ( not previous states ). For instance it would be impossible for a MLP to predict the stpe respons of a system as the input is constat and so will be the output.
Do you feed to the network some kind of previous state too?
Or i'm missing something?

[code2love]

+1 Same question